Truncation logic refactoring

[mateuszkupper]

Instead of specifying a truncation method returning a tensor.Box for all optyx (core) diagrams, a generic truncation method is given for optyx.core.diagrams.Box which uses _truncation_specification which specifies the mapping between input basis vectors to output basis vectors. Only such _truncation_specification method needs to be implemented for all boxes inheriting from optyx.core.diagrams.Box.

Example for W (white triangle):

    def truncation_specification(
        self,
        inp: Tuple[int, ...],
        max_output_dims: Tuple[int, ...] = None
    ) -> Iterable[BasisTransition]:

        for config in occupation_numbers(inp[0], self.n_legs):
            if all(
                config[i] < max_output_dims[i] for i in range(self.n_legs)
            ):
                yield BasisTransition(
                    out=tuple(config),
                    amp=multinomial(config) ** 0.5
                )

Example for Add:

    def truncation_specification(
        self,
        inp: Tuple[int, ...],
        max_output_dims: Tuple[int, ...] = None
    ) -> Iterable[BasisTransition]:
        n = sum(inp)
        if n < int(max_output_dims[0]):
            yield BasisTransition(out=(n,), amp=1.0)

For each input vector (a tuple with a photon number config), we return an output basis vector (photon number configs) with corresponding value which then can be used to construct an array for a tensor. This is done through a named tuple BasisTransition with params out and amp.

Not all current boxes implement this method. For some, truncation is overridden directly (especially when we can give a more efficient implementation) - like for ZBox.

This PR simplifies the truncation logic by moving the common tensor/array building code into a shared method.